This article proposes an alternative approach of Value-at-Risk (VaR) estimation. Financial assets are known to have irregular return patterns; not only the volatility but also the distribution functions themselves may vary with time. Therefore, traditional time-series models of VaR estimation assuming constant and specific distribution are often unreliable. The study addresses the issue and employs the nonparametric kernel estimator technique directly on the tail distributions of financial assets to produce VaR estimates. Various key methodologies of VaR estimation are briefly discussed and compared. The empirical study utilizing a sample of stocks and stock indices for almost 14 years data shows that the proposed approach outperforms other existing methods.